Adaptive Bit and Power Allocation for Indoor Wireless Multicarrier ...

4 downloads 22 Views 267KB Size Report
Furthermore, the benefits of power allocation are noticeable at low signal-to-noise ratios. I. Introduction. The development of high speed wireless net- works over ...

Proceedings of the 15th International Conference on Wireless Communications, (Calgary, AB, Canada), July 2003

Adaptive Bit and Power Allocation for Indoor Wireless Multicarrier Systems Alexander M. Wyglinski

Peter Kabal

Fabrice Labeau

Department of Electrical & Computer Engineering McGill University, 3480 University St. Montr´eal, QC, Canada H3A 2A7 Abstract We propose an adaptive bit and power allocation algorithm for indoor wireless systems employing multicarrier modulation. The proposed scheme maximizes throughput via an incremental allocation algorithm while operating under a maximum mean bit error constraint. Unlike other bit allocation algorithms, which allocate a continuous distribution of bits followed by quantization, the proposed algorithm allocates a discrete distribution of bits. Moreover, the proposed algorithm employs a stricter subband power constraint to limit the interference to other users and satisfy government regulatory requirements, unlike other algorithms which only employ a total power constraint. Finally, the assumption of flat subchannels is dropped, thus a subcarrier minimum mean-squared error equalizer is applied to the case of orthogonal frequency division multiplexing systems employing a cyclic prefix. The performance of the proposed system is evaluated in terms of throughput and bit allocation and compared with an IEEE 802.11acompliant system. The results show that the proposed system outperforms the IEEE 802.11a-compliant system when transmitting at lower signal-to-noise ratios. Furthermore, the benefits of power allocation are noticeable at low signal-to-noise ratios.

I. Introduction The development of high speed wireless networks over the past several years has resulted in the implementation of systems which are capable of transmitting at data rates of up to 54 megabits per second (Mb/s) [1]. Although much work has gone into making these systems reliable, robust in fading channels, and spectrally efficient when transmitting at high data rates, several problems still need to be addressed. In this paper we approach the problem of a frequency-selective fadThis research was partially funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) and Le Fonds de Recherche sur la Nature et les Technologies du Qu´ ebec.

ing channel by using adaptive bit and power allocation in multicarrier systems. In multicarrier modulation, data is transmitted in parallel subcarriers at a lower data rate than the serial input and output of the system. This effectively transforms the frequency selective fading channel into a collection of flat fading subchannels. An efficient version of multicarrier modulation is orthogonal frequency division multiplexing (OFDM), which has no intersymbol interference (ISI) when a sufficiently long cyclic prefix is used. As a result, many wireless local area network (WLAN) standards, such as IEEE 802.11a [1] and HIPERLAN/2 [2], use OFDM systems at the core of their design. Conventional wireless OFDM systems uses a fixed signal constellation size across all subcarriers. Their overall error probabilities are dominated by the subcarriers with the worst performance. One solution to this problem is adaptive bit and power allocation, where the signal constellation size and power distribution across all the subcarriers vary according to the estimated channel conditions in order to minimize the overall error probability. In Kalet [3], Chow et al. [4], Fischer & Huber [5], Hughes-Hartog [6], and Leke & Cioffi [7], the power and bit allocation are optimized in order to either achieve a maximum bit rate for a given probability of error or to minimize the probability of error given a target bit rate; the power level of each subcarrier sums up to a constant total power while the bit allocation can be non-integer with no maximum limit on the constellation size. On the other hand, in the work by Schmidt & Kammeyer [8], Czylwik [9], and Keller & Hanzo [10], adaptive modulation is used, where the subcarriers are adaptively modulated with a fixed number of signal constellation sizes. Although these methods either offer low complexity or near-optimal bit allocations, none offer a balance between these two criteria. Furthermore, none of these algorithms impose practical constraints on the power allocation that meet regulatory requirements. In this paper, we pro-

II. Multicarrier Modulation The general setup for an adaptive multicarrier modulation (MCM) system is shown in Fig. 1. An MCM system divides the input symbol stream into N parallel streams, each having a larger symbol period and lower bandwidth relative to the input. In the case of adaptive MCM, the high data rate input, x(n), is split into N streams with different data rates and employing different modulation schemes. Each of these parallel streams is then used to modulate a carrier using a basis function g (k) (n), 1 ≤ k ≤ N . The modulated streams are summed together and transmitted across the channel, where they experience the effects of multipath propagation as well as noise. The received signal, r(n), is decomposed into the N subcarriers using basis functions f (k) (n), 1 ≤ k ≤ N . The subcarrier equalizers w (k) (n), 1 ≤ k ≤ N , are then applied to each of the received parallel streams and the result is demodulated and multiplexed together to form the received symbol stream. Two examples of MCM systems are orthogonal frequency division multiplexing (OFDM) and filter bank multicarrier (FB-MC) systems. In OFDM systems, the discrete Fourier transform (DFT) and the inverse DFT (IDFT) are employed as basis functions. Although efficiently implemented using the fast Fourier transform (FFT)

x( n )

Serial-to-parallel / Adaptive Subcarrier Modulation

x

x

x

(1)

(n)

(2)

( n)

(N )

( n)

(1)

( n)

( 2)

( n)

N

g

N

g

N

g

Σ

r ( n)

h (n) v ( n)

(N )

( n)

(a) Transmitter with Channel

f r ( n)

f

f

(1)

( n)

N

w

(1)

( n)

x

(2)

( n)

(N)

( n)

(2)

( n)

N

w

(N)

( n)

N

w

x

x

(1)

( n)

( 2)

( n)

(N )

( n)

Parallel-to-serial / Adaptive Subcarrier Demodulation

pose an adaptive bit and power allocation algorithm for wireless multicarrier systems in order to enhance system throughput while satisfying a mean bit error rate constraint. The bit allocation algorithm is performed incrementally while the power allocation satisfies a subband power constraint. To enhance system performance, especially when the number of subcarriers is insufficient to result in flat subchannels, we employ minimum mean squared error equalizers to each subcarrier and adaptively allocate taps to them based on channel conditions. This paper is organized as follows. In Section II, an overview of multicarrier systems is presented, especially with details on orthogonal frequency division multiplexing (OFDM) systems employing cyclic prefixes. The proposed bit and power allocation algorithms are outlined in Section III. In Section IV, the design of the optimal subcarrier MMSE equalizers for OFDM systems employing cyclic prefixes and a description of the equalizer tap allocation algorithm are presented. Section V presents the results of the proposed bit allocation algorithm, bit and power allocation algorithm, and equalizer tap allocation algorithm. Finally, a summary of this work with some concluding remarks are included in Sections VI.

x ( n)

(a) Receiver

Fig. 1 A schematic of the adaptive wireless multicarrier data transmission system

and the inverse FFT (IFFT), the basis functions in OFDM systems possess poor spectral selectivity, thus cyclic prefixes are employed to remedy the situation. In FB-MC systems, the basis functions can be designed to be more spectrally selective than the OFDM basis functions, thus removing the need for a cyclic prefix. This type of MCM system and the application of adaptive allocation was studied in [11]. III. Bit & Power Allocation Algorithms In this section, the proposed bit and power allocation algorithms and its advantages over other allocations algorithms are presented. Before continuing, we briefly perform a literature survey to define the current state-of-the-art. A. Existing Bit & Power Allocation Algorithms One of the first bit and power allocation algorithms for multicarrier systems was developed by Hughes-Hartog [6]. In this algorithm, the incremental energy required to transmit one additional bit is computed for each of the subcarriers. The subcarrier that requires the least incremental energy is then allocated that bit and the process is repeated as long as either the total power or aggregate bit error rate constraints are not violated. The allocation algorithm of Chow, Cioffi, and Bingham [4] makes use of an approximation for the Shannon channel capacity to compute the bit allocation across all the subcarriers. Specifically, they calculate the number of bits for subcarrier i using   SNR(i) 1 (1) b(i) = log2 1 + 2 Γ where Γ is the SNR gap, representing how far the

system is from achieving the Shannon capacity rectangular 64-QAM (M = 64) is given by [13]     for a specific BER [12], and SNR(i) is the SNR 1 3γ of subcarrier i. Assuming equal energy across √ Q PM = 4 1 − M −1 M all used subcarriers, the value for system perfor   (3)   mance margin, γmargin , is iteratively sought af1 3γ √ Q · 1 − 1 − ter until either the target bit rate is satisfied or M −1 M the maximum number of iterations have been performed, in which case the allocation is forced to where the Q-function is defined as Z ∞ the target bit rate using a suboptimal loop. Fi2 1 Q(x) = √ e−t /2 dt nally, the transmission power levels are adjusted 2π x in order to achieve the same BER per used subcarrier. and γ is the subcarrier signal-to-noise ratio. To In the allocation algorithm by Fischer and Hu- obtain the probability of bit error from Eq. (3), ber [5], the objective is to distribute bits and use the approximation Pb ≈ PM / log2 (M ). Using an incremental bit allocation algorithm, power across the subcarriers in order to minimize the error probability on each subcarrier. Using the signal constellation configuration for the subthe closed-form expressions for the symbol error carriers given BERThres can be determined via probabilities for QAM modulation, the algorithm the following algorithm: iteratively distributes the bits and power until the 1. probability of error on all subcarriers are equivalent, subject to a total rate and total power constraint. P_{1-MHz}/3 The allocation algorithm presented by Leke and Cioffi [7] assigns energy to different subcarriers in order to maximize the data rate for a given margin. First, a sort and search is performed in order to find which subcarriers should be left on while others shut off. Second, the energy is distributed either equally or via water-filling. Finally, the bits allocated to each subcarrier is calculated using the approximation for the Shannon channel capacity used in [4]. B. Proposed Bit Allocation Algorithm The guiding principle behind the proposed bit allocation algorithm is to ensure the mean BER is below a specified threshold, BERThres , while maximizing throughput. The proposed algorithm redistributes the bits across all the subcarriers while constrained by the BERThres criterion and a subcarrier bit allocation constraint. This algorithm is an improvement over other algorithms since it does not allocate a continuous distribution of bits which is then quantized [3–5, 7].

Initialize the algorithm by setting the modulation scheme of all the subcarriers to 64QAM and the subcarrier power levels to P1–MHz /4, where P1–MHz is the subband power constraint over any 1 MHz bandwidth in the U-NII band [14].

2. Determine the probability of bit error for all the subcarriers given the instantaneous subcarrier SNR values by using the closed form expressions for the probability of bit error. 3. If the mean BER, BERMean , is less than BERThres , the current configuration is kept and the algorithm is terminated. Otherwise, we proceed to Step 4. 4. Search for the subcarrier with the worst BER and check to see if it is BPSK-modulated. If it is, null the subcarrier and proceed to Step 5. Otherwise, reduce its constellation size, flag the subcarrier, and proceed to Step 6. 5. Readjust the power levels of the subcarrier adjacent to the recently nulled subcarrier, ensuring that the new power levels satisfy the 1 MHz bandwidth power constraint, and flag the subcarriers (see Section III-C for details).

6. Recompute the subcarrier BER values of all To compute the probability of bit error for flagged subcarriers and return to Step 3. all subcarriers, given that the channel conditions In the next subsection, the power allocation alare known at the transmitter, closed-form expresgorithm that operates in tandem with the bit alsions are employed. For instance, the probability location algorithm is described. of bit error for BPSK is given by [13] C. Proposed Power Allocation Algorithm p  2γ (2) Pb,BPSK (γ) = Q Unlike the total power constraint used by most allocation algorithms [3–7, 9], a strict subband while the probability of symbol error for QPSK power constraint (such as the regulatory require(M = 4), rectangular 16-QAM (M = 16), and ments of the FCC [14]) is employed in this work

to reduce interference with other users. In applications such as wireless local area networks, the frequency bands of operation are usually unlicensed and users are non-cooperative. For these, power constraints are essential. The subcarrier power allocation scheme operates by allocating to the non-nulled subcarrier i a power level of l+2 l+3 X Pi = P1–MHz − Pk (4)

subcarrier as  (k) xn−L+1,n = x(k) (n)

···

x(k) (n − L + 1)

T

(6) where xT is the transpose of x. Using the same technique employed in [11] and neglecting the cyclic prefix for now, we treat the FFT and IFFT blocks as Discrete Fourier Transform (DFT) and inverse DFT (IDFT) filterbanks, (k) respectively. Therefore, the signal xn−L+1,n is k=l upsampled by N , filtered by the k th synthek6=i (k) where P1–MHz is the subband power constraint sis filter gn−P +1,n , a channel impulse response (k) over any 1 MHz bandwidth. For the 5.15–5.25 hn−S+1,n , and the k th analysis filter fn−P +1,n , GHz U-NII lower band [14], P1–MHz is equal to before being downsampled by N and equalized 2.5 mW. For instance, since four consecutive sub- by w(k) n−Q+1,n . carriers constitute 1 MHz in IEEE 802.11a, to Filtering is performed in this analysis by usdetermine l use l+2 ( l+3 ) ing convolution matrices. Therefore, we can repX (k) (5) resent gn−P +1,n as an (N Q + P + S − 2) × Pk l = arg max (i−3)≤l≤i k=l (N Q + 2P + S − 3) convolution matrix (i-2)

Suggest Documents